Demand Forecasting Using Coupling Of Machine Learning And Time Series Models For The Automotive After Market Sector

Demand forecasting is considered a crucial component for manufacturing companies, as it directly impacts planning and revenue. This research aims to forecast the demand for automotive products using historical demand history data along with external factors that influence demand. The current approach for generating forecast models takes into account various time-series and machine learning algorithms. The motive was to understand the demand at least granular material number tagged with sales and manufacturing plant offices. A combination of 31792 unique stock-keeping units (SKU) was identified,representing the 37 sales office, 27 plants, 1525 materials. A bag of 23 time-series and machine learning algorithms was used individually over monthly demand history during the past twenty-four months (Jan 2018 – April 2020) and forecasted for one year (May 2020 – April 2021). The exploratory data analysis and feature engineering algorithms were implemented over a complete raw dataset along with data cleaning and characteristics monitoring. The best performingalgorithm is selected for the future forecast by mutual comparison of all individual algorithms mean absolute percentage error as well as ensembled outcomes from time-series and machine-learning algorithms. The best performing model (time-series, machine learning, ensemble) was selected as per the error and accuracy matrix to forecast the demand for the individual sku’s.